2020-03-26T16:20:53Z
Acceptance in OA@INAF
þÿRevisiting the relationship between 6 ¼m and 2-10 keV continuum luminosities of
AGN
Title
Mateos, S.; Carrera, F. J.; Alonso-Herrero, A.; Rovilos, E.; Hernán-Caballero, A.; et
al.
Authors
10.1093/mnras/stv299
DOI
http://hdl.handle.net/20.500.12386/23608
Handle
MONTHLY NOTICES OF THE ROYAL ASTRONOMICAL SOCIETY
Journal
449
Number
Revisiting the relationship between 6
µ
m and 2–10 keV continuum
luminosities of AGN
S. Mateos,
1‹F. J. Carrera,
1A. Alonso-Herrero,
1†
E. Rovilos,
2A. Hern´an-Caballero,
1X. Barcons,
1A. Blain,
3A. Caccianiga,
4R. Della Ceca
4and P. Severgnini
41Instituto de F´ısica de Cantabria (CSIC-Universidad de Cantabria), E-39005 Santander, Spain
2Institute for Astronomy, Astrophysics, Space Applications & Remote Sensing, National Observatory of Athens, 15236 Palaia Penteli, Greece 3Physics and Astronomy, University of Leicester, University Road, Leicester LE1 7RH, UK
4INAF–Osservatorio Astronomico di Brera, via Brera 28, I-20121 Milano, Italy
Accepted 2015 February 11. Received 2015 February 10; in original form 2014 October 2
A B S T R A C T
We have determined the relation between the AGN luminosities at rest-frame 6µm associated with the dusty torus emission and at 2–10 keV energies using a complete, X-ray-flux-limited sample of 232 AGN drawn from the Bright Ultra-hard XMM-Newton Survey. The objects have intrinsic X-ray luminosities between 1042 and 1046erg s−1 and redshifts from 0.05 to
2.8. The rest-frame 6µm luminosities were computed using data from the Wide-field Infrared Survey Explorer and are based on a spectral energy distribution decomposition into AGN and galaxy emission. The best-fitting relationship for the full sample is consistent with being linear, L6µm∝ L0.99±0.032−10 keV, with intrinsic scatter, log L6µm∼ 0.35 dex. The L6µm/L2−10 keV
luminosity ratio is largely independent of the line-of-sight X-ray absorption. Assuming a constant X-ray bolometric correction, the fraction of AGN bolometric luminosity reprocessed in the mid-IR decreases weakly, if at all, with the AGN luminosity, a finding at odds with simple receding torus models. Type 2 AGN have redder mid-IR continua at rest-frame wavelengths <12 µm and are overall ∼1.3–2 times fainter at 6 µm than type 1 AGN at a given X-ray luminosity. Regardless of whether type 1 and type 2 AGN have the same or different nuclear dusty toroidal structures, our results imply that the AGN emission at rest-frame 6µm is not isotropic due to self-absorption in the dusty torus, as predicted by AGN torus models. Thus, AGN surveys at rest-frame∼6 µm are subject to modest dust obscuration biases.
Key words: galaxies: active – quasars: supermassive black holes – infrared: galaxies – X-rays:
galaxies.
1 I N T R O D U C T I O N
In the standard unified model of active galactic nuclei (AGN), the observed differences between AGN subclasses are solely explained as the result of an orientation effect with respect to an optically and geometrically thick distribution of molecular gas and dust on tens of parsecs scales, frequently assumed to have a toroidal geometry (Antonucci1993; Urry & Padovani1995). This material, hereafter referred to as the dusty torus, absorbs the AGN bolometric lumi-nosity output produced at rest-frame UV/optical and X-ray wave-lengths in the accretion disc/corona and re-emits it mainly in the mid-infrared (mid-IR) regime (rest-frame wavelengths∼5–30 µm). Thus, it is clear that the mid-IR spectral range offers a unique op-portunity to study the obscured accretion phenomenon. Surveys
E-mail:mateos@ifca.unican.es
†Augusto G. Linares Senior Research Fellow.
conducted using data from Spitzer’s Infrared Array Camera (IRAC) have identified a population of highly obscured AGN whose number density is comparable to that observed for unobscured AGN. Many of these objects might be hidden at optical and X-ray wavelengths by Compton-thick material (line-of-sight rest-frame gas column densities NH> 1.5 × 1024cm−2; Mart´ınez-Sansigre et al.2005).
Uncovering the mid-IR thermal continuum associated with the AGN dusty torus emission can be a difficult task, due to contam-ination from the AGN host galaxies. High (subarcsecond) spatial resolution observations can isolate the AGN emission from other contaminants on nearby objects with favourable geometries. Al-ternatively, mid-IR spectral decomposition can provide reliable es-timates of the AGN dusty torus contribution to the infrared light (e.g. Ballo et al.2014). Studies such as these have found a tight correlation between the intrinsic luminosity, traced by hard (2– 10 keV) X-rays, and reprocessed mid-IR emission of AGN over more than three orders of magnitude in luminosity (Lutz et al.2004; Ramos Almeida et al.2007; Horst et al.2008; Gandhi et al.2009;
2015 The Authors
at INAF Brera Milano (Osservatorio Astronomico di Brera) on April 28, 2015
http://mnras.oxfordjournals.org/
Levenson et al. 2009; Asmus et al. 2011, 2014; Mason et al.
2012). The relationship between hard X-ray and mid-IR luminosi-ties seems to be identical for AGN optically classified as type 1 and type 2, even for Compton-thick AGN. If the hard X-ray emission is a good tracer of both the intrinsic power of AGN and the heat-ing source of the dusty torus, then these studies provide strong observational support for the mid-IR light being a reliable proxy for the AGN intrinsic power, regardless of any obscuration. Un-fortunately, the AGN samples involved in these studies are highly incomplete and biased towards infrared-bright objects. Recent stud-ies of local AGN detected with the Swift Burst Alert Telescope and AKARI all-sky surveys have shown a tight correlation be-tween the hard X-ray (>10 keV energies) and monochromatic mid-IR (9 and 18µm) luminosities and that unabsorbed and ab-sorbed AGN follow the same correlation (Ichikawa et al. 2012; Matsuta et al.2012).
The sensitive all-sky infrared survey conducted with the
Wide-field Infrared Survey Explorer (WISE; Wright et al.2010) at 3.4, 4.6, 12 and 22 µm will dramatically increase our AGN census over a significant fraction of the age of the Universe improving greatly our understanding of the obscured accretion phenomenon in luminous systems. It has already been demonstrated that mid-IR colour-based selection techniques with WISE, targeting the charac-teristic red power-law thermal continuum emission from the AGN dusty torus (fν∝ να with α≤ −0.5; Alonso-Herrero et al.2006),
are highly reliable and effective at uncovering both unobscured and obscured luminous AGN (e.g. Assef et al.2010,2013; Mateos et al.2012,2013; Stern et al.2012; Yan et al.2013). As demon-strated in Mateos et al. (2013), WISE can potentially uncover many Compton-thick AGN missed at other wavelengths, at least up to
z ∼ 1. Despite such progress, the physical properties of the AGN
revealed with WISE remain poorly constrained.
In this paper, we investigate how well we can isolate the AGN emission associated with the dusty torus in the mid-IR regime us-ing the WISE broad-band photometric data and, more importantly, whether the mid-IR continuum luminosities derived in this way are a reliable isotropic proxy for the intrinsic luminosity of AGN. To do so we have conducted a UV-to-mid-IR spectral energy distribution (SED) decomposition into AGN and galaxy components. Then we have studied the correspondence between the AGN luminosities at rest-frame 6µm, corrected for the accretion disc and host galaxy emission, and at 2–10 keV energies over more than three orders of magnitude in AGN luminosity. We have used the final data re-lease of the WISE survey (AllWISE; Cutri et al.2013) and a large uniform complete sample of 232 X-ray-selected AGN drawn from the Bright Ultra-hard XMM-Newton Survey (BUXS). The objects have z in the range 0.05–2.8, and intrinsic (absorption-corrected) 2–10 keV X-ray luminosities between 1042 and 1046erg s−1. The
X-ray luminosities, corrected for absorption, were computed from a detailed analysis of the XMM–Newton X-ray spectroscopic data available for all objects.
This paper is structured as follows. Section 2 describes the AGN sample and the multiwavelength data used in this study. In Sec-tion 3 we describe the SED fitting approach used to isolate the AGN mid-IR emission, the adopted technique to compute rest-frame 6µm luminosities and discuss the effects associated with contamination from the AGN host galaxies. In Section 4 we present the relationship between 6µm and 2–10 keV continuum luminosi-ties. The main results are discussed in Section 5 and summarized in Section 6. Throughout this paper, errors are 68 per cent confidence for a single parameter, and we assume M = 0.3, = 0.7 and H0= 70 km s−1Mpc−1.
2 AG N S A M P L E D E S C R I P T I O N
Our AGN sample was drawn from the wide-angle BUXS (Mateos et al.2012). BUXS is a large, complete, flux-limited sample of X-ray bright (f4.5−10 keV> 6 × 10−14erg s−1cm−2) AGN detected with
the XMM–Newton observatory at 4.5–10 keV energies. The survey is based on 381 high Galactic latitude (|b| > 20◦) XMM–Newton observations having good quality for serendipitous source detection that were used to derive extragalactic source count distributions at intermediate X-ray fluxes (Mateos et al. 2008). BUXS contains 255 AGN, after removal of Galactic stars and known BL Lacs (<3 per cent), detected over a total sky area of 44.43 deg2.
2.1 UV/optical spectroscopic follow-up and AGN classification All the sources in BUXS are in the area covered by the Sloan Digi-tal Sky Survey (SDSS) imaging survey Data Release 7 in the ugriz bands (Abazajian et al.2009). To identify the optical counterparts, we used the likelihood-ratio estimator cross-matching algorithm of Pineau et al. (2011)1(see Mateos et al.2012for details). All sources
but two have detections in SDSS. For the remaining two sources, we identified their optical counterparts from our own imaging cam-paign.
Optical spectroscopic identifications and classifications are cur-rently available for all but five objects. Such high identification completeness (98 per cent) guarantees that our study will not suffer from biases associated with low optical identification rates. The data have been collected from the literature (mainly from SDSS) and as part of our extensive optical follow-up identification programme (full details will be presented in a forthcoming paper).
Throughout this paper, objects with detected UV/optical emis-sion line velocity widths≥1500 km s−1and intermediate Seyfert types 1–1.5 are classified as type 1 AGN (143) while Seyfert types 1.8, 1.9 and 2 and those objects with a galaxy spectrum with no emission lines are classified as type 2 AGN (107). In this way, we have a more uniform optical spectroscopic classification over the
z range sampled by BUXS. For example, Seyfert 1.9s would be
classified as type 2 AGN at z 0.2, when the Hα line at 6563 Å is redshifted from our optical spectra.2We have checked that the
results of our study do not change if we instead classify Seyferts 1.8 and 1.9 as type 1 AGN.
2.2 Broad-band UV-to-mid-IR photometric data
As discussed below, one of the main goals of our study is to deter-mine the rest-frame 6µm mid-IR continuum emitted by the AGN dusty torus. To do so we have decomposed the rest-frame UV-to-mid-IR SEDs of our AGN into AGN and host galaxy components. To build the SEDs, we have collected all the available photometric data from SDSS, the Two Micron All Sky Survey (2MASS), the UKIRT Infrared Deep Sky Survey (UKIDSS) and WISE.
We have used SDSSMODELMAG magnitudes as they provide re-liable estimates of the fluxes of extended objects while, for point sources, they are indistinguishable from the standard point spread function (PSF) magnitudes. We converted SDSS magnitudes and errors into flux densities following the SDSS recommendations.3
1The code is available fromhttp://saada.u-strasbg.fr/docs/fxp/plugin/
2The optical spectra from our follow-up identification programme do not
typically extend to wavelengths longer than∼8000 Å.
3http://www.sdss.org/dr7/algorithms/fluxcal.html
at INAF Brera Milano (Osservatorio Astronomico di Brera) on April 28, 2015
http://mnras.oxfordjournals.org/
A 2 per cent error was added in quadrature to the catalogued flux errors to account for the uncertainties in the zero-points.
We have used the cross-matching algorithm of Pineau et al. (2011) to identify the WISE mid-IR counterparts of our AGN (see Mateos et al.2012for details). All sources but six (five type 1 AGN and one type 2 AGN) have detections with SNR≥ 2 at the WISE wave-lengths of 3.4, 4.6 and 12µm in AllWISE. The mean X-ray-mid-IR separation is2 arcsec. Only 10 objects have a non-zero AllWISE photometric quality flag in at least one of the WISE bands indicating that the fluxes may be contaminated due to proximity to an image artefact (cc f lags; Cutri et al.2013). After a careful visual check of the WISE images and SEDs, we found no issues in the WISE fluxes for any of the sources so we decided to keep them in our sample.
We computed flux densities in the WISE bands using either ellip-tical aperture magnitudes for the objects flagged as extended (six in total) or profile fitting photometry for the sources not spatially resolved. We used the magnitude zero-points on the Vega system
Fν(iso)= 309.124, 171.641, 30.988 and 8.346 Jy for 3.4, 4.6, 12
and 22µm, respectively (Wright et al.2010). They correspond to a spectrum fν∝ να with α= −1. We added 1.5 per cent
uncertain-ties to the catalogued flux errors in all bands to account for the overall systematic uncertainty from the Vega spectrum in the flux zero-points. To account for discrepancies between the red and blue calibrators used for the conversion from magnitudes to janskys, we added an additional 10 per cent uncertainty to the 12 and 22µm catalogued fluxes (Wright et al.2010).
At the stage of the project when we searched for the near-IR photometric information available for all the AGN in BUXS, we had already accurate source positions for the optical counterparts of all the objects in the sample. Therefore, to identify the near-IR counterparts of our AGN, we cross-correlated the coordinates of their optical counterparts with the 2MASS catalogues of point and extended sources (Jarrett et al. 2000; Cutri et al. 2003) and the UKIDSS Data Release 10 catalogues (Lawrence et al.2007) using a simple position matching with a search radius of 2 arcsec. In total 166 objects (∼71 per cent) have detections in the near-IR. The mean optical–infrared separation is <1 arcsec. We have checked that increasing the search radius does not produce any additional matches. From UKIDSS we used Petrosian magnitudes for extended sources and the default aperture magnitudes for point sources. As for SDSS, we added a 2 per cent error in quadrature to the catalogued flux errors to account for uncertainties in the zero-points. 2MASS and UKIDSS magnitudes were converted into fluxes using the zero-points Fν(iso)= 1594, 1024 and 666.7 Jy for the 2MASS JHKs
(Cohen, Wheaton & Megeath 2003) and Fν(iso)= 2232, 2026,
1530, 1019 and 631 Jy for the UKIDSS ZYJHK (Hewett et al.
2006).
We note that if we use the WISE magnitude zero-points in the Vega system corresponding to the spectral shapes assumed to convert magnitudes to flux densities in UKIDSS (∼ν0) and 2MASS (∼ν−2),
the differences in the derived flux densities in the WISE bands would be lower than 1 per cent at 3.4, 4.6 and 22µm and 9 per cent at 12 µm, all below the systematic uncertainties in the catalogued fluxes.
2.3 X-ray spectral properties
Here we give a brief summary of the analysis conducted to determine the X-ray properties of our AGN. Full details on the X-ray spectral extraction procedure, analysis of the data and X-ray properties of
the full sample of BUXS AGN will be presented in a forthcoming paper.
We have good quality XMM–Newton X-ray spectra in the ob-served energy range from 0.25 to 10 keV for all the AGN in this study. The median (mean) background-subtracted number of counts in the 0.25–10 keV EPIC spectra (MOS+pn) is 1443 (3127), where the counts range from 52 to 75 000. The X-ray spectroscopic analy-sis was conducted with theXSPECpackage (v12; Arnaud1996). For each object we fitted the data with a combination of different mod-els to determine the shape of the direct and scattered continuum components (modelled with power laws) and rest-frame line-of-sight X-ray absorption (considering both full coverage and partial coverage of the continuum source). All models include the Galac-tic absorption for each source using column densities taken from Dickey & Lockman (1990). We have also accounted for the ‘soft excess’, i.e. any emission at energies1 keV in excess of the ex-trapolation of the primary continuum, an almost ubiquitous feature in the X-ray spectra of AGN (e.g. Turner et al.1997; Porquet et al.
2004; Guainazzi, Matt & Perola2005; Piconcelli et al.2005; Ma-teos et al.2010; Scott, Stewart & Mateos2012; Winter et al.2012) using thermal and blackbody models. To accept the detection of a model component, we used the F-test with a significance threshold of 95 per cent.
If X-ray absorption was not detected, we computed 1σ upper lim-its using the best-fitting model parameters. Though BUXS includes many objects with X-ray absorption close to being Compton thick, when taking into account the uncertainties in the best-fitting rest-frame column densities, all X-ray absorption values are consistent with being in the Compton-thin regime (NH< 1.5 × 1024cm−2).
Thus, based on the existing X-ray data we do not have any securely confirmed Compton-thick AGN. The X-ray luminosities used in this paper have been determined from the best-fitting models in the rest-frame 2–10 keV energy band and are corrected for X-ray absorption. Examples of our X-ray spectral modelling are shown in Fig.2.
2.4 AGN sample selection
For the study presented here, we have selected the 232 AGN in BUXS with absorption-corrected X-ray luminosities≥1042erg s−1
and WISE detections in the three shortest wavelength bands (3.4, 4.6 and 12µm) with SNR ≥ 2. Out of these, 137 objects are type 1 AGN (122 Seyfert 1s, 15 Seyfert 1.5s) and 95 type 2 AGN (3 Seyfert 1.8s, 14 Seyfert 1.9s and 78 Seyfert 2s). The X-ray luminosity limit has been imposed to reduce to a minimum the effects of host galaxy contamination in the mid-IR regime by increasing the contrast of the AGN contribution.
The selected AGN have 2–10 keV X-ray luminosities between 1042and 1046erg s−1and z in the range 0.05–2.8. TableA1lists the
optical spectroscopic classification and intrinsic X-ray luminosity for all the AGN used in our study while Table1summarizes their main properties. The distribution of their z and 2–10 keV luminosi-ties is shown in Fig.1.
3 AG N M I D - I R C O N T I N U U M L U M I N O S I T I E S Our study aims to investigate how well we can isolate the mid-IR emission associated with the AGN dusty torus using WISE broad-band photometric data, and whether the mid-IR AGN luminosities determined in this way are a reliable isotropic proxy of the AGN in-trinsic power. To do so we have studied the correspondence between
at INAF Brera Milano (Osservatorio Astronomico di Brera) on April 28, 2015
http://mnras.oxfordjournals.org/
Table 1. Summary of the main properties of the AGN used in this study.
Sample N log L2–10 keV log L2–10 keV z z Nabs
(erg s−1) (erg s−1)
(1) (2) (3) (4) (5) (6) (7)
All 232 42.12–46.00 43.94(43.94) 0.056–2.860 0.558(0.396) 122(52.6 per cent)
Type 1 137 42.32–46.00 44.23(44.38) 0.061–2.860 0.713(0.646) 37(27.0 per cent)
Type 2 95 42.12–45.19 43.52(43.41) 0.056–1.266 0.334(0.240) 85(89.5 per cent)
Notes. Column 1: sample; Column 2: number of sources; Column 3: 2–10 keV X-ray luminosity range (in logarithmic units). Column 4: median (mean) X-ray luminosity (in logarithmic units); Column 5: redshift range; Column 6: median (mean) redshift; Column 7: number (fraction) of objects with detected X-ray absorption.
Figure 1. Hard X-ray luminosity versus z for the AGN in BUXS selected for this study.
the rest-frame 6µm and 2–10 keV X-ray continuum luminosities for our complete, flux-limited sample of X-ray-selected AGN.
We have used rest-frame 6µm monochromatic luminosities as a tracer of the AGN dusty torus emission. As such emission can be well approximated by a power law (see e.g. Neugebauer et al.
1979; Elvis et al.1994; Richards et al.2006; Assef et al.2010), to compute the 6µm luminosities we have used a simple linear interpo-lation/extrapolation in log–log space between the WISE-catalogued fluxes at 4.6 and 12µm. Section 3.1 describes in detail how we cor-rected these luminosities for both host galaxy contamination and for the emission from the AGN accretion disc.
We have not used the WISE 3.4µm fluxes because, as shown in Section 3.1, they are severely contaminated by starlight. Moreover, we have not used the WISE 22µm fluxes because, as this survey is much shallower than the ones conducted at shorter wavelengths, many of our objects are not detected. Indeed, out of the 232 AGN selected for this study just 168 (72.4 per cent) are detected with SNR≥ 2 at 22 µm.
3.1 SED decomposition into the AGN and their hosts
Even at the high AGN luminosities involved in our study, the 4.6 and 12µm fluxes (rest-frame ∼1.5–4 µm at z ∼ 2) might be contami-nated by the emission from the AGN host galaxies. To isolate the AGN component and derive the 6µm luminosity associated with the dusty torus, we have decomposed the rest-frame UV-to-mid-IR SEDs of our objects into AGN and galaxy emission. We used the best-fitting results to correct the 6µm luminosities for contami-nation from the AGN host galaxies and accretion disc emission. We stress that the main goal of the SED fitting analysis is to prop-erly model all the continuum components in the observed SEDs
to ensure a reliable determination of the dusty torus emission. We have used the fitting code SEd Analysis using BAyesian Statistics (SEABASS)4described in detail in Rovilos et al. (2014). The program
uses the maximum likelihood method and a Markov chain Monte Carlo (MCMC) sampling technique to determine both the optimum combination of SED templates and the associated uncertainties.
3.1.1 SED components
We fitted the SEDs with three model components to reproduce the emission from the AGN accretion disc, the dusty torus and the stellar emission of the hosts.
To describe the UV/optical emission from the accretion disc, we have used the average type 1 quasar SED of Richards et al. (2006) at λ < 0.7µm and a power law λfλ∝ λ−1at longer wavelengths.
To redden the accretion disc, we have used the Gordon & Clayton (1998) Small Magellanic Cloud extinction law at λ < 3300 Å, as it best represents the nuclear reddening of quasars (Hopkins et al.
2004), while at λ > 3300 Å we used the Galactic extinction law from Cardelli, Clayton & Mathis (1989). In both cases we assume
RV= 3.1.
We have characterized the mid-IR emission from the AGN dusty torus with the template library of Silva, Maiolino & Granato (2004). The library includes four templates divided according to the amount of X-ray absorption. These templates were generated by combining the nuclear mid-IR observations of a large sample of Seyfert galaxies and interpolation using the smooth-density dust emission model of Granato & Danese (1994). Of these we have used the Seyfert 1 tem-plate and the two Seyfert 2 temtem-plates corresponding to X-ray column densities in the Compton-thin regime: 1022< N
H< 1023cm−2and
1023< N
H< 1024cm−2. We have not used either the optical
clas-sification or best-fitting X-ray column densities to choose amongst torus templates, i.e. we have considered all three torus templates to fit the SEDs of our type 1 and type 2 AGN.
Finally, we have reproduced the emission from the stellar popu-lation of the AGN hosts at rest-frame optical–near-IR wavelengths with the stellar population synthesis models of Bruzual & Charlot (2003). We used a library of 75 stellar templates with solar metal-licity and a Chabrier initial mass function (Chabrier2003). We have used 10 exponentially decaying star formation histories with char-acteristic times τ = 0.1–30 Gyr and a model with constant star formation, and a set of ages in the range 0.1–13 Gyr (see Lusso et al.2013for a similar approach). We have checked that none of our fits require such long (>10 Gyr) and probably unphysical ages. We reddened the templates using the Calzetti et al. (2000) dust extinction law and a range of E(B− V)galfrom 0 to 2.
4http://astro.dur.ac.uk/∼erovilos/SEABASs/
at INAF Brera Milano (Osservatorio Astronomico di Brera) on April 28, 2015
http://mnras.oxfordjournals.org/
To avoid degeneracies between the AGN and galaxy templates, we have ignored any possible contamination from dust heated by star formation in the AGN hosts. As demonstrated in Section 3.3, at the high AGN luminosities involved in our study and rest-frame wavelengths sampled with the WISE 4.6 and 12µm surveys, con-tamination due to star formation has a negligible effect on our results. Nevertheless, as the star formation emission rapidly in-creases at rest-frame wavelengths 8 µm, contamination of the
WISE 22µm fluxes might not be negligible at the lowest redshifts
(and AGN luminosities). To minimize the impact of such effect, we have treated all 22µm detections (168 in total) as upper limits.
In the following subsections, we discuss in more detail the SED fitting procedure and summarize the main results. In the AGN with signatures of the broad-line region in their rest-frame UV/optical spectrum (i.e. Seyfert types 1 to 1.9) the contamination from the accretion disc at UV/optical and near-IR wavelengths is not neg-ligible. Therefore, we have carried out the SED fits following a different approach for AGN with detected UV/optical broad emis-sion lines (i.e. where we have a relatively unobscured view of the BLR and accretion disc; Section 3.1.2) and for AGN without de-tected UV/optical broad emission lines (i.e. where the accretion disc emission is highly obscured; Section 3.1.3). Nevertheless, we remind the reader that, to have a more uniform optical spectroscopic classification over the z range sampled by BUXS, throughout the paper we have classified Seyfert 1.8–1.9 (17 in total) as type 2 AGN (see Section 2.1).
3.1.2 SED fitting of AGN with detected UV/optical broad emission lines
For the type 1 AGN (137) and intermediate Seyfert types 1.8– 1.9 (17), we have used a range of reddening values for the accretion disc, E(B− V )disc, between 0 and 0.65 with a step E(B − V )disc= 0.01. As indicated in Caccianiga et al. (2008), the
separation between optical type 1 and type 2 AGN corresponds to
AV∼ 2 mag, or E(B − V) ∼0.65 assuming a Galactic standard
con-version. To take into account any intrinsic dispersion in the SEDs of our AGN, the normalizations of the disc and torus components were left free to vary. The SED fitting was conducted both with and without the stellar component. We accepted the detection of the stellar component if the χ2difference between models, χ2, was
higher than 6.17 (2σ confidence for two parameters).5Otherwise,
we did not apply any correction for host galaxy contamination to the observed WISE fluxes.
We have found a median (mean) nuclear extinction E(B−
V )disc = 0.05 (0.11), with 68 per cent of the sources having E(B − V )discbelow 0.1. Considering only those sources showing
evidence for both X-ray absorption (see Section 2.3) and UV/optical reddening (49 in total), we found that the great majority (88 per cent) have dust-to-gas ratios E(B− V)/NHlower than Galactic by a
fac-tor varying from 1.5 to more than 100 in a few cases, in good agreement with Maiolino et al. (2001). In 42 objects (38 with
L2–10 keV> 1044erg s−1) adding the stellar component did not
im-prove the quality of the fit above the selected threshold. The median (mean) contribution of the accretion disc at rest-frame 6µm, relative to the best-fitting torus model, is 11.8+7.6−6.1(12.9± 7.8) per cent and has no significant luminosity dependence based on the Spearman rank test (correlation coefficient ρ= −0.12 and probability to reject
5The relationship between χ2 and the log-likelihood difference is
χ2= −2(lnL) (see appendix in Rovilos et al.2014for details).
the null hypothesis 86.7 per cent). The median uncertainty associ-ated with the correction for the accretion disc emission determined from the SED fits is 13 per cent. Such uncertainties were added in quadrature to our 6µm luminosity estimates.
3.1.3 SED fitting of AGN without detected UV/optical broad emission lines
In the 78 AGN without detected broad emission lines in their rest-frame UV/optical spectra, the accretion disc emission is highly ob-scured and the observed fluxes at rest-frame UV-to-near-IR wave-lengths are dominated by the AGN host galaxies. Nevertheless, for AGN with nuclear reddening levels AV∼ 2–5 mag, the accretion
disc emission at near-IR wavelengths (∼1–2 µm) could still be rel-evant. As for these objects we do not have a direct view of the accretion disc, we cannot constrain both the extinction and nor-malization of such a component. The approach we followed was to fix the normalization of the accretion disc relative to the torus component to the median value obtained for the AGN with detected UV/optical broad lines. We then created a library of reddened accre-tion disc templates using a range of E(B− V )discvalues between
0.65 and 10 with a step E(B− V )disc= 0.15. These templates
were added to each of the three torus templates used for the SED fitting. We repeated the fitting process, fixing the accretion disc nor-malization relative to the torus component to the median−1σ and median+1σ values, and used the best-fitting results to determine the uncertainties in the correction of the 6µm luminosities for the accretion disc emission.
The best-fitting E(B− V )discvalues have a nearly uniform
dis-tribution between the chosen limits with a median (mean) value of E(B− V )disc= 6.22 (5.95). We stress, however, that the
mea-sured nuclear extinctions should be regarded as less accurate than for objects where we have a direct view of the accretion disc. The median (mean) contribution of the reddened accretion disc at rest-frame 6µm, relative to the best-fitting torus model, is 6.5+3.9−2.6 (6.9± 3.3) per cent. The median uncertainty associated with the correction for the accretion disc emission from the SED fits is 46 per cent. Although as pointed out previously, the extinction val-ues are less accurate, we found that 67 per cent of the sources have dust-to-gas ratios lower than Galactic by a factor that varies from ∼1.5 to 90, again in very good agreement with Maiolino et al. (2001).
Examples of the SED fitting procedure for type 1 and type 2 AGN are shown in Fig.2. The contamination from the AGN host galaxies at rest-frame near-IR and mid-IR wavelengths is discussed in Sections 3.2 and 3.3 and summarized in Figs3and4, respectively. In Table2, we report the median contribution from the AGN hosts and the reddened accretion disc at rest-frame 6µm for type 1 and type 2 AGN at different X-ray luminosities.
3.2 Stellar contamination in the rest-frame near-IR and mid-IR
We have corrected our rest-frame 6 µm luminosities for the Rayleigh–Jeans tail of the emission of the different underlying stel-lar populations of the AGN host galaxies as follows. First, we have computed the observed fluxes in the 4.6 and 12µm passbands asso-ciated with starlight emission using the best-fitting stellar templates. To accurately reproduce the measurement process with the wide bandpasses of the WISE filters, we have convolved the best-fitting stellar SEDs with the WISE relative spectral response (RSR) filter
at INAF Brera Milano (Osservatorio Astronomico di Brera) on April 28, 2015
http://mnras.oxfordjournals.org/
Figure 2. Examples of the modelling of the XMM–Newton EPIC X-ray spectroscopic data and SED decomposition for AGN classified as type 1 (left plots)
and type 2 (right plots), respectively. Top: the X-ray spectra for 2XMMJ095908.3+024309 are best modelled with a simple unabsorbed power-law model
(broad-band photon index = 1.88 ± 0.02; solid lines). For 2XMMJ221813.9+001624 the best-fitting model consists of an absorbed power-law ( = 1.77+0.39−0.37
and NH= 40.4+8.0−7.1, where the column density is in units of 1022cm−2) plus a low-energy scattered power-law with same spectral index as that of the direct
absorbed component (dotted lines). Bottom: the figures clearly illustrate that, at the shortest wavelength bands of WISE, the emission from the AGN host galaxies can dominate over that of the dusty torus especially in type 2 AGN where the extinction of the AGN emission is systematically higher.
curves. Then, we have determined the luminosity at 6µm associated with starlight emission through linear interpolation/extrapolation in log–log space. The median uncertainty associated with the correc-tion for stellar emission determined from the SED fits is 19 per cent. These were added in quadrature to our 6µm luminosity uncertain-ties.
Fig.3shows the relative contribution from starlight to the fluxes measured with WISE at 3.4, 4.6 and 12µm as a function of AGN luminosity. In the cases where the emission from the AGN hosts was not detected with sufficient significance (χ2< 6.17), we assigned
a zero value to the fractions. For AGN optically classified as type 2, the starlight contamination to the WISE fluxes at 3.4 and 4.6µm is important across the full luminosity range sampled by our study. This is expected as dust extinction of the AGN emission is more important in type 2 objects. In addition, in flux-limited surveys, such as BUXS, the most luminous objects are detected at higher z and the WISE passbands are sampling shorter rest-frame wavelengths where the starlight emission peaks. As dust extinction of the AGN emission is less severe in type 1 objects, the starlight contamination to their mid-IR fluxes is significantly smaller than that for type 2 AGN but still non-negligible, especially at 3.4µm (Kotilainen et al.
1992; Alonso-Herrero, Ward & Kotilainen1996; Alonso-Herrero
et al.2003). Only the WISE fluxes at 12µm are dominated by the AGN at all luminosities, at least up to z∼ 2. An alternative way to visualize the dependence of the level of stellar contamination to the WISE fluxes on the AGN luminosity and optical class is representing the distribution of WISE colours for our sources. If we adopt the mid-IR colour definition of Mateos et al. (2012) (i.e. log(f4.6µm/f3.4µm) versus log(f12µm/f4.6µm) flux density ratios), the
most contaminated objects should be located in the bottom-right part of the colour–colour diagram. Indeed, as we see in fig. 6 in Mateos et al. (2012) the less luminous objects, in particular those classified as type 2 AGN, have overall the bluest log(f4.6µm/f3.4µm)
mid-IR colours.
Fig.4illustrates the correction for starlight emission to our rest-frame 6µm luminosities computed as indicated above. The corre-sponding values are reported in Table2.
3.3 Contamination from star formation in the rest-frame mid-IR
As indicated in Section 3.1.1, we have not corrected our 6 µm luminosities for the emission from dust in star-forming regions in the AGN hosts. For example, polycyclic aromatic hydrocarbon (PAH)
at INAF Brera Milano (Osservatorio Astronomico di Brera) on April 28, 2015
http://mnras.oxfordjournals.org/
Figure 3. Contribution from the AGN host galaxies to the mid-IR fluxes
measured with WISE at 3.4, 4.6 and 12µm (observed frame). Large filled
symbols and vertical error bars indicate the median and 16 and 84 percentiles (68 per cent enclosed, equivalent to 1σ ) of the distribution of values in different luminosity intervals indicated with the horizontal error bars. In the cases where the lower limit of the error bar reached zero, the upper limit corresponds to the 68 percentile (i.e. smallest interval that encloses the 68 per cent probability).
Figure 4. Host galaxy contribution to the rest-frame 6µm luminosities as a function of the AGN luminosity. Large filled symbols and vertical error bars indicate the median and the 16 and 84 percentiles (68 per cent enclosed, equivalent to 1σ ) of the distribution of values in different luminosity intervals indicated with the horizontal error bars. In the cases where the lower limit of the error bar reached zero, the upper limit corresponds to the 68 percentile (i.e. smallest interval that encloses the 68 per cent probability).
features can probe star formation in the AGN hosts over time-scales of tens of millions of years (e.g. Peeters, Spoon & Tielens2004). Nevertheless, it is expected that any PAH redshifted into the WISE-observed bands superimposed on the strong AGN continuum would be smoothed out due to the wide bandpasses of the filters, especially for the extremely wide 12µm filter. To demonstrate that this is the case, we have determined upper limits on the level of contamination from star formation to our 6µm luminosities. To do so, we have used the main-sequence and starburst galaxy templates from Elbaz et al. (2011). For each object in our sample, we have determined the upper limit in the normalization of the star-forming templates assuming that the hosts of our AGN are luminous infrared galaxies (LIRGs) with L8−1000µm≤ 6 × 1011L and star formation rates
(SFRs) of∼100 M yr−1based on the Kennicutt (1998) relation. The less than equal sign indicates the additional constraint that none of the fluxes determined from each template in the WISE bands (taking into account the filter RSRs) should be above the observed values. Then, we computed the 6µm luminosities from linear interpolation/extrapolation of the 4.6 and 12µm fluxes in log– log space as we did for the original data. Following this approach, we found that the contribution from star formation at 6µm is well below 15 per cent for 83.3 per cent of the AGN with L2–10 keVin the
range 1042–1043erg s−1 and for >99 per cent of the sources with L2–10 keV≥ 1043erg s−1.
If we assume instead that the hosts of our AGN with
z > 0.5 are ultraluminous infrared galaxies (ULIRGs) with L8−1000µm ≤ 6 × 1012L (very IR luminous galaxies become
more numerous at high z) and SFRs∼1000 Myr−1, we find that the contamination from star formation at rest-frame 6µm is less than 15 per cent for 89.1 per cent of these sources. Table2lists our (very conservative) median upper limits on the level of contamina-tion from star formacontamina-tion at rest-frame 6µm as a function of AGN class and luminosity.
For comparison, the median upper limit on the star formation contamination to the WISE 12µm passband is 18 per cent at AGN luminosities 1042< L
2–10 keV< 1043erg s−1, and less than 4 per cent
at higher luminosities. At the rest-frame wavelengths sampled with
at INAF Brera Milano (Osservatorio Astronomico di Brera) on April 28, 2015
http://mnras.oxfordjournals.org/
Table 2. Contribution from the reddened accretion disc and the AGN hosts to the mid-IR at rest-frame 6µm.
Sample N log L2–10 keV f6stµm f6discµm f6disc+stµm f6sbµm
(erg s−1) (per cent) (per cent) (per cent) (per cent)
(1) (2) (3) (4) (5) (6) (7) All 36 42–43 20.76+22.81−15.88 3.66+11.57−2.47 28.13+17.94−11.23 7.93+8.92−4.79 Type 1 12 42–43 10.95+16.97−8.13 15.23+14.96−5.72 27.15+18.46−10.25 7.52+9.32−5.38 Type 2 24 42–43 26.02+19.95−14.94 2.94+2.32−2.16 29.25+17.13−12.37 9.53+7.72−5.60 All 86 43–44 7.81+8.93−5.10 7.39+7.18−4.07 17.96+8.73−6.36 2.54+3.20−1.89 Type 1 39 43–44 5.09+6.78−3.99 10.79+6.09−5.23 18.32+8.55−8.54 1.65+2.56−1.18 Type 2 47 43–44 11.57+5.91−6.12 5.94+3.83−3.18 17.41+8.26−5.63 3.19+3.22−1.91 All 88 44–45 2.12+2.30−2.12 10.82+5.25−5.87 13.84+7.08−4.34 0.37+0.93−0.25(3.72+9.29−2.46) Type 1 65 44–45 1.40+2.67−1.40 11.40+6.74−5.22 14.18+6.88−4.68 0.32+0.62−0.21(3.25+6.17−2.14) Type 2 23 44–45 4.40+5.18−2.88 8.73+3.60−4.11 13.17+4.10−2.91 0.63+1.33−0.44(6.27+13.30−4.45) Type 1 21 45–46 0.23+0.83−0.23 11.33+4.23−5.54 12.45+6.94−6.20 0.01+0.05−0.01(0.11+0.65−0.06)
Notes. Column 1: sample; Column 2: number of sources; Column 3: 2–10 keV X-ray luminosity range (in logarithmic units). Columns 4 and 5: median contribution from the AGN hosts stellar emission and the reddened
accretion disc at rest-frame 6µm, respectively. Column 6: median of the sum of the contributions from the AGN
hosts and the reddened accretion disc at rest-frame 6µm. Column 7: (very conservative) median upper limits
for the level of contamination from star formation at rest-frame 6µm assuming that the AGN hosts are LIRGs
or ULIRGs (values in brackets; see Section 3.3 for details). The error bars represent the 16 and 84 percentiles. In the cases where the lower limit of the error bar reached zero, the upper limit corresponds to the 68 percentile (i.e. smallest interval that encloses the 68 per cent probability). Please note that to compute the numbers we have
not included the only type 2 AGN in the sample with L2–10 keV> 1045erg s−1.
the WISE 3.4 and 4.6µm passbands, the contamination from star formation should be negligible.
We note that to compute the above numbers we have used the full sample of objects, i.e. including those for which we have only upper limits on their 22µm fluxes. From the comparison of the 12 µm flux distributions and (f22µm/f12µm) flux density ratios (using 1σ upper
limits for f22µm if not detected), we have found that the objects
in BUXS that have escaped detection at 22µm are ∼4–5 times fainter at 12µm than those detected at 22 µm, but they also have on average lower (∼0.8–0.9 times smaller) mid-IR flux density ratios. This implies even lower levels of contamination from dust in star-forming regions in the objects not detected with WISE at 22µm.
3.4 Reliability of rest-frame 6µm luminosities for z > 1 AGN
It is well known that there is some curvature in the IR continuum emission of AGN at rest-frame wavelengths close to the∼1 µm inflection point between the UV and near-IR bumps (Elvis et al.
1994; Richards et al.2006). This means that our linear interpola-tion/extrapolation approach, based on the WISE fluxes at 4.6 and 12µm, might be overestimating the 6 µm luminosities for objects at z > 1, when the WISE 4.6µm band is sampling rest-frame wavelengths2 µm. To investigate such effect, we have compared our derived 6µm luminosities with those obtained using the WISE fluxes at 12 and 22µm. We have used for this comparison all AGN in BUXS with z > 1 and detection with SNR≥ 2 at 22 µm (19 out of 37 objects with z > 1). We have found a median (mean) difference in the logarithm of the 6µm luminosities of just −0.02 (−0.02). Although the numbers involved in the comparison are small, these results show that at up to z∼ 2, rest-frame 6 µm AGN luminosi-ties can be robustly determined using the WISE 4.6 and 12µm fluxes alone. In addition, we confirm that contamination from dust emission in star-forming regions to the WISE 22µm fluxes is not important.
4 C O R R E L AT I O N O F L6µm A N D L2–10 keV
L U M I N O S I T I E S
4.1 Linear regression technique
From now own, we will refer to the mid-IR monochromatic lumi-nosities at rest-frame 6µm as L6µm while L2–10 keV are the
rest-frame 2–10 keV X-ray luminosities corrected for X-ray absorption (see Section 2.3). To determine the relationship between L6µmand L2–10 keV, we have performed a linear regression in log–log space
using two different approaches: the least-squares χ2minimization
(Press et al.1992) and the Bayesian maximum likelihood method proposed by Kelly (2007, hereafterK07).6For theK07technique,
the data are fitted using the Metropolis–Hastings MCMC algo-rithm sampler. We have adopted uniform prior distributions for the regression parameters. The best-fitting parameters and their corre-sponding uncertainties have been determined by taking the median and the 16 and 84 percentiles (68 per cent enclosed, equivalent to 1σ ) from the posterior probability distributions of the model pa-rameters using 104iterations from the MCMC sampler. The two
regression techniques account for measurement errors in both the independent and dependent variables. However, since theK07 ap-proach determines simultaneously both the regression parameters and the intrinsic scatter,7it should provide more reliable estimates
of the parameters, their associated uncertainties and of the strength of the linear correlation. Therefore, from now on we will refer to the results obtained with theK07technique, but present those obtained from the least-squares technique for comparison with results in the literature.
6http://idlastro.gsfc.nasa.gov/ftp/pro/math/linmix_err.pro
7If x and y are the independent and dependent variables, respectively, the
scatter in y at fixed x that would have been measured if there were no errors in y.
at INAF Brera Milano (Osservatorio Astronomico di Brera) on April 28, 2015
http://mnras.oxfordjournals.org/
Table 3. Summary of the results of the linear regression analysis.
Sample Method N ρ (p-value) τ (p-value) σobs σint α β γ
(1) (2) (3) (4) (5) (6) (7) (8) (9) (10) All LS 232 0.90 (<10−5) 0.42 (<10−5) – 0.44± 0.01 1.01± 0.01 – All Bayesian 232 0.918+0.010−0.012 – 0.38 0.350−0.018+0.020 0.302+0.026−0.025 0.986+0.030−0.030 2.003+0.120−0.113 Type 1 LS 137 0.88 (<0.0013) 0.40 (<10−5) – 0.52± 0.01 0.89± 0.01 – Type 1 Bayesian 137 0.904+0.017−0.020 – 0.40 0.353−0.022+0.028 0.377+0.033−0.034 0.941+0.042−0.041 2.372+0.197−0.175 Type 2 LS 95 0.88 (<10−5) 0.45 (<10−5) – 0.24± 0.01 1.00± 0.01 – Type 2 Bayesian 95 0.889+0.023−0.027 – 0.40 0.325−0.027+0.031 0.180+0.048−0.047 0.926+0.058−0.059 1.513+0.180−0.158
Notes. Column 1: sample; Column 2: linear regression technique. LS: least squares; Column 3: number of objects; Column 4: linear correlation coefficient. For least squares, the Spearman rank and the associated probability that the correlation is entirely caused by distance based on our Monte Carlo simulations; Column 5: Kendall’s partial correlation coefficient and probability that the correlation
is entirely caused by distance based on our Monte Carlo simulations; Column 6: observed scatter in L6µmabout the regression line;
Column 7: intrinsic scatter about the regression line. At a given X-ray luminosity, σintdenotes the standard deviation in the mid-IR
luminosity; Columns 8 and 9: intercept and slope of the regression, respectively (equation 1); Column 10: constant in linear space (equation 2). Uncertainties are 1σ .
The linear regression techniques used here assume that the mea-surement errors are Gaussian. However, this is not the case for either the X-ray luminosity measurements or the uncertainties associated with the correction of the mid-IR luminosities for the accretion disc and starlight emission. To overcome this problem, for each data point with luminosity L and associated uncertainties L− and
L+(all values in logarithmic units) we have computed a Gaussian function with mean L and dispersion σ such that its integral from
L − L− to L+ L+is equal to the enclosed probability given by the fits (68.3 per cent for the X-ray luminosity uncertainties and 95.4 per cent for the mid-IR luminosity uncertainties derived from
SEABASS). We have used the 1σ uncertainties derived in this way in
our regression analysis.
To determine the L6µm–L2–10 keVrelationship, we have conducted
the linear regression analysis treating L2–10 keVas the independent
variable and L6µmas the dependent variable. Our decision is based
on the fact that hard X-rays, produced at distances much closer to the central SMBH (subparsec scales), are a more direct tracer of the AGN intrinsic power than the mid-IR emission, originating from the dusty torus at tens of parsec scales.8We have assumed a linear
relation of the form log L6µm 1044 = α + β × log L2−10 keV 1044 , (1)
where the luminosities are in units of erg s−1 and α and β are the correlation parameters. The luminosities have been normalized at L2–10 keV = 1044erg s−1, roughly the mean AGN power of our
sample.
8From a statistical point of view we note that, in the presence of
intrin-sic scatter, the best-fitting slopes of forward and inverse regressions (i.e. switching the dependent and independent variables) can be very different. To circumvent such problem, both variables are often treated symmetrically
using the bisector line (e.g. Isobe et al.1990). However, if the dependent
variable of the linear regression analysis depends on the parameter used to select the sample (although our AGN sample is X-ray flux limited, we stress
that we are only using objects from BUXS with L2–10 keV> 1042erg s−1),
the truncation of the data set will artificially flatten the measured slope due to the increased loss of objects with low X-ray luminosities at low mid-IR luminosities (K07). Thus, the conventional forward fit relation, i.e. fitting
L6µmat a given L2–10 keV, should provide the more accurate estimate of
the existing relationship between the mid-IR and X-ray luminosities for our AGN.
Equation (1) can be written as L6µm 1044 = γ × L2−10 keV 1044 β , (2)
where we have computed γ (∼10α) and its 1σ uncertainties as the median and the 16 and 84 percentiles of the probability density distributions of γ . The latter were computed with a Monte Carlo sampling of the posterior probability distributions of α.
4.2 The L6µm–L2–10 keVrelation
Table3summarizes the results of our regression analysis. The best-fitting linear relations for the full sample and the type 1 and type 2 AGN obtained with theK07 technique are illustrated in Fig.5. TableA1lists the X-ray and 6µm luminosities of our AGN.
Although theK07technique determines the strength of the cor-relation between L6µm and L2–10 keV, we have also computed the
Spearman rank-order correlation coefficient (ρ) for comparison. Both the Spearman rank and the Bayesian correlation coefficients confirm the tight correlation between mid-IR and X-ray luminosi-ties over more than three decades in AGN luminosity (from 1042to
1046 erg s−1).
To investigate the role of distance in our luminosity–luminosity correlation, we have also performed a partial Kendall τ correla-tion test taking into account the common dependence of L6µmand L2–10 keV on the distance. To compute the significance of the
mea-sured Spearman ρ and Kendall τ correlation coefficients for our sample, we have conducted a ‘scrambling test’ (e.g. Merloni et al.
2006; Bianchi et al.2009). The idea behind this test is as follows: for each source, we keep its X-ray luminosity and z. We then assign them 6µm fluxes from the sample at random with replacement (bootstrap) and they are converted to a mid-IR luminosity using the z of the source. In this way, we are removing any intrinsic correlation between L6µmand L2–10 keVthat is not associated with
distance effects. We have repeated this Monte Carlo test 105times
calculating each time the Spearman ρ and Kendall τ correlation co-efficients. The simulations were conducted for both the full sample and for the type 1 and type 2 AGN samples. The results are listed in Table3and are illustrated in Fig.6for the full sample (similar results were obtained for the type 1 and type 2 AGN). The values of the correlation coefficients obtained for the original data set are well above the distribution of those for the simulated samples. Our
at INAF Brera Milano (Osservatorio Astronomico di Brera) on April 28, 2015
http://mnras.oxfordjournals.org/
Figure 5. Rest-frame 6µm monochromatic luminosities against 2–10 keV intrinsic (i.e. corrected for absorption) X-ray luminosities. The 6 µm luminosities have been corrected for the accretion disc emission and contamination from the AGN host galaxies as indicated in Section 3. The best-fitting linear relations
in log–log space derived with theK07Bayesian approach for the full sample and the type 1 and type 2 AGN are indicated with solid, dotted and dashed lines,
respectively. We compare our results with the relation derived for type 1 AGN by Lutz et al. (2004, grey shading).
Figure 6. Results of the Monte Carlo simulations conducted to determine
the correlation between L6µmand L2–10 keV(solid histograms). The large
histogram represents the distribution of the Kendall partial correlation co-efficient (where distance effects have been taken into account) while the inset plot illustrates the results for the Spearman correlation coefficient. The vertical dashed lines represent the correlation coefficients obtained for the original sample.
analysis shows that the probability that the measured correlation between L6µmand L2–10 keVis entirely due to the range of distances
in our sample is almost always <10−5(see Table3).
At a given X-ray luminosity, the standard deviation in 6µm luminosities is very similar for both type 1 (∼0.353 dex) and type 2 (∼0.325 dex) AGN and the slopes of the linear regression relations are compatible, within the uncertainties. Despite this, type 1 AGN are overall∼1.3–2 times brighter than type 2 AGN at 6 µm at a given X-ray luminosity, albeit with a large overlap between both populations.
The shaded area in Fig.5shows the relation (and 1σ dispersion) for type 1 AGN from Lutz et al. (2004). We have compared the mid-IR/X-ray luminosity ratio distributions of our AGN and the 42 objects in Lutz et al. (2004) with L2–10 keV> 1042erg s−1, detections
at 6µm and estimates of the intrinsic X-ray luminosity. According to the two-sample Kolmogorov–Smirnov (KS) test, the probability of rejecting the null hypothesis (the two samples are drawn from the same parent population) is 61.3 per cent. We can therefore conclude that the two distributions are consistent with each other. We have also compared the L12µm/L2–10 keVluminosity ratio distributions of
our AGN and those from Gandhi et al. (2009) with X-ray luminosi-ties >1042erg s−1(37 AGN), where L
12µmare the monochromatic
luminosities in νLνunits at rest-frame 12µm. To do the comparison, we computed L12µm(corrected for the accretion disc emission and
contamination from the AGN host galaxies) for each of our AGN as indicated in Section 3. According to the KS test, the distributions of
L12µm/L2–10 keVare consistent with each other (47.3 per cent
prob-ability of rejecting the null hypothesis). Therefore, our results are compatible with published mid-IR:X-ray relations for local AGN (e.g. Krabbe, B¨oker & Maiolino2001; Lutz et al.2004; Gandhi et al.2009; Levenson et al.2009; Asmus et al.2014). None the less, they do not support the luminosity-dependent relationship de-rived for the type 1 AGN in the Chandra COSMOS (C-COSMOS) and Chandra Deep Field South (CDFS) surveys suggested by Fiore et al. (2009).
We have demonstrated that at AGN luminosities
L2–10 keV > 1042erg s−1, the mid-IR emission detected with WISE at rest-frame wavelengths ∼6 µm is dominated by hot
dust heated by the AGN. Therefore, it is possible to indirectly infer the intrinsic 2–10 keV X-ray luminosities of AGN with data from WISE by inverting equations (1) and (2) after correcting the 6 µm luminosities for contamination from the AGN hosts
at INAF Brera Milano (Osservatorio Astronomico di Brera) on April 28, 2015
http://mnras.oxfordjournals.org/
and the accretion disc. Moreover, we have shown that such corrections are not negligible, at a ∼30 per cent level for AGN with 1042 < L
2–10 keV < 1043erg s−1 and at a ∼12–18 per cent
level for AGN with 1043 < L
2–10 keV< 1046erg s−1. Based on the
full sample, X-ray luminosities can be determined with a root mean square (rms) error of 0.376 dex. The corresponding rms values for type 1 and type 2 AGN are 0.404 dex and 0.355 dex, respectively.9
5 D I S C U S S I O N
5.1 Dispersion of the L6µm–L2–10 keVrelation
Several factors can contribute to the measured scatter of the L6µm
versus L2–10 keVrelationship, such as the inherent large dispersion
in the individual SED shapes associated with the varying proper-ties of the material responsible for the X-ray and mid-IR emission and AGN variability effects (the X-ray and mid-IR observations have been taken several years apart). As X-rays are produced in a compact region close to the central SMBH, they trace more di-rectly changes in the AGN bolometric radiative output than the mid-IR.
AGN torus models with both smooth and clumpy dust distri-butions predict anisotropic dust emission at rest-frame 6µm (e.g. Pier & Krolik1992; Fritz, Franceschini & Hatziminaoglou2006; Nenkova et al.2008a,b). Although smooth-density models typically predict higher levels of mid-IR anisotropy (e.g.∼an order of mag-nitude from face-on to edge-on views at rest-frame 6µm; Pier & Krolik1992), it has recently been shown that the anisotropy level depends mainly on the model assumptions and not so much on the adopted dust morphology (i.e. smooth or clumpy; Feltre et al.
2012). Nevertheless, if the mid-IR emission is not isotropic at the rest-frame wavelengths involved in our study, such effect should contribute to the measured intrinsic dispersion in the L6µmversus L2–10 keVrelationship.
We note that, if contamination from star formation were the main contributor to the observed spread, this would imply a contamina-tion to the 6µm luminosities greater than 1–10−σint ∼ 56 per cent on average. This is unlikely to be the case as demonstrated in Sec-tion 3.3. Moreover, as our sample consists of AGN with absorpSec-tion in the Compton-thin regime, uncertainties associated with the cor-rection of the X-ray luminosities for intrinsic X-ray absorption are generally small given the good quality X-ray spectroscopic data available for all sources. This is supported by the fact that the in-trinsic spread of L6µmat a given L2–10 keVis comparable for (mostly
unabsorbed) type 1 and (highly absorbed) type 2 AGN.
5.2 A luminosity-independent dusty torus?
Several works in the literature claim that the fraction of AGN bolo-metric luminosity that is reprocessed in the mid-IR increases at a lower rate than the AGN intrinsic bolometric power (e.g. Bar-cons et al.1995; Maiolino et al. 2007; Treister, Krolik & Dulle-mond2008; McKernan et al.2009; Mor, Netzer & Elitzur2009; Calderone, Sbarrato & Ghisellini2012). These results have been
9To determine the rms, we have used the 6µm luminosity range where we
can compute the dispersion in X-ray luminosities, log L6µm> 42.7 erg s−1
for the full sample and type 2 AGN and log L6µm> 42.8 erg s−1for type
1 AGN.
interpreted in terms of a luminosity-dependent unification model, where the properties of the AGN dusty torus, such as the geomet-rical covering factor, vary with luminosity (e.g. Lawrence1991; Simpson2005; Della Ceca et al.2008). Other studies find, how-ever, either weak or no luminosity dependence (Gandhi et al.2009; Levenson et al.2009; Roseboom et al.2013).
We have determined a relationship between L6µmand L2–10 keV
for both type 1 and type 2 AGN that is consistent with being linear. Furthermore, the average rate of change of L6µm with respect to L2–10 keVis similar in the two AGN classes. Fig.7shows the
distribu-tions of log(L6µm/L2–10 keV) for type 1 and type 2 AGN at different
X-ray luminosities. Table4lists the mean and median of these dis-tributions. We find a mild variation, if at all, of log(L6µm/L2−10 keV)
with luminosity, in any case well within the mutual dispersions.
In this comparison, we are implicitly assuming that the 2–10 keV intrinsic luminosity is a reliable proxy for the AGN bolometric radiative power and that the hard X-ray bolometric correction is constant across the luminosity range of our sample. Recent works in the literature suggest that the X-ray emission of AGN is likely weakly anisotropic, in the sense that type 2 AGN appear intrinsically fainter than type 1 AGN at X-ray energies (e.g. Diamond-Stanic, Rieke & Rigby2009; LaMassa et al.2010; Dicken et al.2014; Liu et al.2014and references therein). If this were the case, this should have no effect on the results obtained for our type 1 AGN. On the other hand, taking into account that the amplitude of the X-ray anisotropy does not seem to vary with luminosity (Liu et al.2014), a correction for X-ray anisotropy effects would move the points for the type 2 AGN in Fig.5to the right but, should not change significantly the slope of the correlation between their mid-IR and X-ray luminosities.
We note that if the X-ray bolometric correction increases with AGN luminosity, then the power-law spectral slope of the equiv-alent relation between L6µm and the bolometric luminosity Lbol
would decrease. If we adopt, for example, the luminosity-dependent bolometric correction of Marconi et al. (2004), we would obtain
L6µm∝ L0.750±0.024bol for the full sample of objects (for a similar
luminosity-dependence of the hard X-ray bolometric correction, see also Hopkins, Richards & Hernquist2007). More recent stud-ies in the literature suggest, however, a much weaker dependence of the hard X-ray bolometric correction with AGN luminosity if any (Vasudevan & Fabian2007; Marchese et al.2012; Fanali et al.
2013).
Finally, we do not expect our results to be affected by sample selection biases. On the one hand, as indicated in Section 2.2, only six sources in BUXS do not have detections in AllWISE with SNR above our detection threshold. Hence, our AGN sample is not bi-ased against infrared-faint sources (these sources would occupy the lower-right corner in Fig.5). On the other hand, spectroscopic follow-up studies of infrared-selected AGN candidates did not find evidence for the existence of a population of AGN with infrared-to-X-ray luminosity ratios significantly higher than those of the AGN revealed in X-ray surveys, at least not for AGN with X-ray lumi-nosities and z in the range sampled with the BUXS survey (e.g. Lacy et al.2013).
Thus, our study shows that, if the fraction of AGN bolometric luminosity that is reprocessed in the mid-IR decreases with the AGN luminosity, the effect is certainly weak in both type 1 and type 2 AGN over the three decades in X-ray luminosity covered with our sample. This finding is at odds with simple receding torus models (e.g. Simpson2005).
at INAF Brera Milano (Osservatorio Astronomico di Brera) on April 28, 2015
http://mnras.oxfordjournals.org/
Figure 7. Distribution of 6µm to X-ray luminosity ratios (in logarithmic units). The vertical lines at the top of the figure indicate the median values of the distributions. The bottom plot also shows the distribution (and its
median) for the type 1 AGN with L2–10 keV> 1045erg s−1.
5.3 Anisotropy of the AGN emission at 6µm
In Section 4.2, we have shown that at a given X-ray luminosity, X-ray-selected type 1 AGN are on average∼1.3–2 times more lu-minous at rest-frame wavelengths∼6 µm than type 2 AGN although with a considerable overlap (see e.g. Ramos Almeida et al.2007;
H¨onig et al.2011; Yang, Wang & Liu2015for similar results). A KS test gives a probability of 2.4 per cent that the distributions of log(L6µm/L2−10 keV) for type 1 and type 2 AGN are drawn from the
same sample. As pointed out in Section 5.2, the AGN X-ray emis-sion is likely weakly anisotropic. If this is the case, a correction for such effect would make the 6µm-to-X-ray luminosity ratios smaller for type 2 AGN and thus the differences in mid-IR luminosities at a given X-ray luminosity between type 1 and type 2 AGN would be even larger than our estimation. To better understand this effect, we show in Fig.8the median rest-frame UV-to-mid-IR SEDs of our AGN in three X-ray luminosity bins before applying any correction for host galaxy contamination. These were computed by first shift-ing the individual SEDs to a common rest-frame wavelength grid. We then converted the fluxes to luminosities in νLν units, and we
combined them in wavelength bins containing at least 15 points. Each value represents the median of the points in the bin while the dispersion is defined with the 16 and 84 percentiles. We did not take into account upper-limits in the computation of the median SEDs hence, as not all AGN were detected with WISE at 22µm, our median SEDs might be somewhat redder than they should at the longest wavelengths (>10µm).
Fig.8also shows the median SEDs of the best-fitting host galaxy stellar emission templates. Although a description of the properties of the host galaxies of our AGN is beyond the scope of this paper, we see that on average there are no significant differences in the near-IR luminosities of the type 1 and type 2 AGN hosts (a proxy for their stellar masses) except at luminosities 1042–1043 erg s−1.
We note, however, that the results obtained at these luminosities should be treated with caution. The number of type 1 AGN is rather small and hence the sample might not be representative of the overall population. Furthermore, at X-ray luminosities from 1042
to 1043erg s−1, the emission from the host galaxies could dilute
the AGN signatures in a significant fraction of objects (50 per cent; see fig. 5 in Caccianiga et al. 2007). Hence, it is likely that at
L2–10 keV < 1043erg s−1 some type 1 AGN in BUXS might have
been misclassified as type 2 AGN.
The dispersion in the bins illustrated in Fig.8reflects not only changes in the intrinsic continuum shape of the different SED com-ponents between objects but also the varying levels of host galaxy contamination and the dispersion in AGN luminosities. Fig.8shows that, even after accounting for the small differences in X-ray lumi-nosities between type 1 and type 2 AGN in the bins, type 2 AGN are overall fainter sources at rest-frame wavelengths4 µm out to at least 12µm than type 1 AGN at a given X-ray luminosity.
To investigate in more detail whether the observed continuum shape of the mid-IR emission associated with the dusty torus in type 1 and type 2 AGN is different, we have compared the distri-butions of the (L6µm/L12µm) luminosity ratios for the two AGN
classes. The 12µm luminosities were determined using linear ex-trapolation in log–log space between the WISE-catalogued fluxes at 4.6 and 12µm and were corrected for both the accretion disc emis-sion and contamination from the AGN host galaxies. The results are presented in Fig.9. Clearly, the mid-IR continuum of type 2 AGN is on average redder10than for type 1 AGN, albeit with a large overlap
between both populations. According to the KS test, the probability of rejecting the null hypothesis (the two samples are drawn from the same parent population) is higher than 99.99 per cent. We note that if we use instead the observed L6µmand L12µmluminosities (i.e.
10The term red (blue) refers to increasing (decreasing) fluxes in νf
νat longer
wavelengths.
at INAF Brera Milano (Osservatorio Astronomico di Brera) on April 28, 2015
http://mnras.oxfordjournals.org/
